Key points
- Microsoft announced Guardian Agents on Monday, a feature that monitors the AI agents a company runs, as part of its Project Perception security product.
- Microsoft Defender now detects up to 60 types of employee-installed agents, including Claude Code, Codex and OpenClaw.
- New controls limit agent access to data and systems, adjust permissions by task, and require human approval when an agent seems to be making risky decisions.
- Microsoft joined Nvidia's Open Agent Safety Platform, which isolates agents and can stop their behavior through hardware, and which more than 100 other companies back.
Why it matters for marketers: Teams letting agents touch customer data, CMS and ad accounts should expect security reviews to demand independent monitoring, permission limits and human approval steps for risky actions.
Microsoft on Monday introduced new security features aimed at the risks posed by AI agents inside companies. The centerpiece is Guardian Agents, which monitors the agents an organization uses. It sits within Project Perception, a product Microsoft introduced in June that was built around the assumption that AI attackers would come from outside.
The framing has changed. Microsoft CEO Satya Nadella has argued that enterprises should stop relying on model providers' safety assurances and instead run separate systems to control agents. Security chief Hayete Gallot wrote that frontier models cannot be trusted, not because they are malicious, but because they are unpredictable and trained to complete tasks rather than follow organizational norms. TechCrunch noted the position is both unexpected, given Microsoft's ties to OpenAI and Anthropic, and commercially convenient for a cloud provider selling security.
For marketers and content teams, the practical point is access. Agents that publish content, edit customer records or run campaigns often hold broad permissions. Microsoft's additions, including detection of agents employees bring in on their own, task-based permission changes and human approval for risky decisions, describe the controls security teams may soon expect around such tools.
Watch how IT and security groups apply these ideas, and whether the Nvidia-led Open Agent Safety Platform gains traction beyond its backers. Now is a reasonable time to list which agents your team uses, what each can reach, and where a person should sign off before an agent acts.
Sources
This brief is our own summary. Read the full reporting at the source.
